Prediction of hepatitis B virus lamivudine resistance based on YMDD sequence data using an artificial neural network model

This Article

Citations


Creative Commons License
Except where otherwise noted, this work is licensed under Creative Commons Attribution-NonCommercial 4.0 International License.
This article, by Kowsar Corp , is licensed under a Creative Commons Attribution License .

Article Information:


Group: 2011
Subgroup: Volume 11, Issue 2, Feb
Date: February 2011
Type: Original Article
Start Page: 108
End Page: 113

Authors:

  • Mehrdad Ravanshad
  • Department of Virology, Faculty of Medical Sciences, Tarbiat Modares University, Tehran, IR.IRan
  • Farzaneh Sabahi
  • Department of Virology, Faculty of Medical Sciences, Tarbiat Modares University, Tehran, IR.IRan
  • Shahab Falahi
  • Department of Microbiology, School of Medicine, Ilam University of Medical Sciences, Ilam, IR.IRan,
  • Azra kenarkoohi
  • Department of Virology, Student Research Committee, Faculty of Medical Sciences, Tarbiat Modares University, Tehran, IR.Iran
  • Samad Amini Bavil Olaee
  • Department of Biotechnology, Pasteur Institute of Iran, Tehran, IR Iran
  • Seyed Younes
  • Hosseini, Department of Virology, Faculty of Medical Sciences, Tarbiat Modares University, Tehran
  • Hossein Riahi
  • Madvar, Water Structures and Engineering Department, Tarbiat Modares University, Tehran
  • Sayad Khanizade
  • Department of Virology, Student Research Committee, Faculty of Medical Sciences, Tarbiat Modares University, Tehran, IR.IRan

      Correspondence:

      Affiliation: Department of Virology, Faculty of Medical Sciences, Tarbiat Modares University
      City, Province: Tehran,
      Country: IR.IRan
      Tel: +98-2182883836
      Fax: +98-2182883836
      E-mail: ravanshad@modares.ac.ir

Abstract:


Background: Hepatitis B virus (HBV) infection is an important health problem worldwide with critical outcomes. The nucleoside analog lamivudine (LMV) is a potent inhibitor of HBV polymerase and impedes HBV replication in patients with chronic hepatitis B. Treatment with LMV for long periods causes the appearance and reproduction of drug-resistant strains, rising to more than 40% after 2 years and to over 50% and 70% after 3 and 4 years, respectively.
Objectives: Artificial neural networks (ANNs) were used to make predictions with regard to resistance phenotypes using biochemical and biophysical features of the YMDD sequence.
Patients and Methods: The study population comprised patients who were intended for surgery in various hospitals in Tehran-Iran. An ACRS-PCR method was performed to distinguish mutations in the YMDD motif of HBV polymerase. In the training and testing stages, these parameters were used to identify the most promising optimal network. The ideal values of RMSE and MAE are zero, and a value near zero indicates better performance. The selection was performed using statistical accuracy measures, such as root mean square error (RMSE), coefficient of determination (R2), and mean absolute error (MAE). The main purpose of this paper was to develop a new method based on ANNs to simulate HBV drug resistance using the physiochemical properties of the YMDD motif and compare its results with multiple regression models.
Results: The results of the MLP in the training stage were 0.8834, 0.07, and 0.09 and 0.8465, 0.160.04 in the testing stage; for the total data, the values were 0.8549, 0.115, and 0.065, respectively. The MLP model predicts lamivudine resistance in HBV better than the MLR model.
Conclusions: The ANN model can be used as an alternative method of predicting the outcome of HBV therapy. In a case study, the proposed model showed vigorous clusterization of predicted and observed drug responses. The current study was designed to develop an algorithm for predicting drug resistance using chemiophysical data with artificially created neural networks. To this end, an intelligent and multidisciplinary program should be developed on the basis of the information to be gained on the essentials of different applications by similar investigations. This program will help design expert neural network architectures for each application automatically.


  • Implication for health policy/practice/research/medical education:
    One of the main obstacles of effective oral antiviral treatment regimens for patients with HBV is viral resistance against oral medications. We suggest reader’s attentions in the field of gastroenterology and liver diseases to the conclusion of this article.
  • Please cite this paper as:
    Ravanshad M, Sabahi F, Falahi S, kenarkoohi A, Amini-Bavil-Olyaee S, Hosseini SY, et al. Prediction of hepatitis B virus lamivudine resistance based on YMDD sequence data using an artificial neural network model. Hepat Mon. 2011;11(2):108-113.

Keywords: Drug resistance;Lamivudine;Neural network models

Manuscript Body:


Background
Infections and cancers that are caused by hepatitis B virus (HBV) are important worldwide health problems with critical outcomes (1). For persons who are chronically infected with HBV, there are two therapeutic approaches that are used to control the infection and its consequences: immunomodulatory agents and antiviral chemotherapy (2). The nucleoside analog lamivudine (LMV) is the optimal therapeutic choice; it inhibits HBV polymerase and slows HBV replication in patients who are chronically infected with hepatitis B (1-5). Drug resistance remains a global public health problem (6), and resistance to LMV is emerging (2). This phenomenon is mediated primarily by mutations in the genes of viruses that alter a drug's interaction with its corresponding target protein (6). Antiviral drug resistance depends on the frequency of viral mutations, the intrinsic mutability of the antiviral target site, and the magnitude and rate of viral replication (5). Typically, mutations in the YMDD motif of the polymerase gene develop after the first 6 months of treatment (7, 8). Long-term therapy with LMV induces the emergence and propagation of drug-resistant isolates, rising to more than 40% after 2 years and to over 50% and 70% after 3 and 4 years, respectively(1, 2, 7). Lamivudine resistance in HBV is a major clinical brake to its long-term use (9).To avoid exacerbation of the infection, the prediction and detection of lamivudine-resistant viruses are vital processes in clinical management (1). Since structural information is available for only a small percentage of proteins, methods for the direct prediction of resistance, based on viral sequences, are highly desired (6). Lamivudine-resistant viruses have a characteristic amino acid substitution in the tyrosine-methionine-aspartate-aspartate (YMDD) motif of its RNA-dependent DNA polymerase. Several studies have reported various mutations that are induced by lamivudine therapy (7, 8, 10, 11). These mutations in the YMDD motif are necessary and sufficient to confer high-level lamivudine resistance (11).It is important to detect the YMDD motif mutations during lamivudine treatment (10). When mutations occur, the configuration of the wild-type YMDD motif becomes altered such that the drug no longer exerts its inhibitory action at that site (2). Numerous techniques have been introduced to monitor HBV drug resistance, such as oligonucleotide chips (12), line probe assay (13), light cycler probe hybridization assay (14), polymerase chain reaction with peptide nucleic acid clamping (15), mass spectrometry of oligonucleotide fragments (16), fluorescence polarization, and sequencing (17). Most of these techniques are accurate but time-consuming, labor-intensive, and difficult to adapt to high-throughput screening. To be used as a clinical evaluation tool and reduce the cost of therapy, methods that distinguish responders from nonresponders and predict outcomes of the treatments must be established (18). Recently, computer-based models have been used by health care providers for management purposes. Neural networks can solve clinical problems based on symptoms and patterns (19). Drug resistance is a complex phenomenon for which several mechanisms are responsible (6). The progress in informatics and its application in decision-making has led to the development of novel artificial intelligence techniques, including artificial neural networks (ANNs) (20).The ANN has been applied in various disciplines of science and technology (21). Statistical learning methods, such as neural networks (22-28), support vector machines (SVMs) (22, 24, 25, 29, 30), and decision rules (22-24, 27, 29, 30), have also shown potential in predicting resistance mutations. An important application of the ANN is the prediction of responses across heterogeneous domains (21). This study performs a novel examination of the use of biochemical and structural information and neural processing in the context of HIV drug resistance (28).ANNs learn by an iterative process that adjusts the strengths of connections, such that the system generates an appropriate result. Notably, data processing by these systems does not require assumptions of how outputs relate to inputs or that inputs be independent (31). Pattern recognition algorithms, including ANNs, have been used widely to analyze biological sequences. Neural learning algorithms, such as back-propagation neural networks (BPNNs), self-organizing maps (SOMs), and recurrent neural networks (RNNs), have been used to analyze protein sequences (32). ANNs allow one to investigate complex, nonlinear relationships. Neural networks are therefore ideally suited for use in drug design (33, 34).

Objectives
ANNs were used to make resistance phenotype predictions from biochemical and biophysical features of the YMDD sequence.

Materials And Methods

Patient and samples
Sera samples were collected from patients who were intended for surgery in various hospitals in Tehran-Iran (35). Patients who were infected with HBV who had not been treated with lamivudine and were negative for HCV and HIV markers were included. YMDD (wild-type pattern), YVDD, and YIDD mutant viral strains were distinguished by an in-house ACRS-PCR assay (36). Eleven samples were sequenced randomly to confirm the ACRS-PCR data.

Artificial Dataset
The main purpose of this paper was to develop a new method based on artificial neural networks to simulate HBV drug resistance and compare its results with multiple regression models. There are several databases of HBV sequences. We used the DDBJ (http: //www.ddbj.nig.ac.jp/) and Expasy databases (http: //www.expasy.ch/) to develop our strategy, because they are used most often in the literature. To develop the model, HBV sequences from GenBank (http://www.ncbi.nlm.nih.gov/sites/entrez) and DDBJ were collected. The Expasy and HIV databases (http://www.hiv.lanl.gov/content/sequence/ENTROPY/entropy_one.html) were used to estimate biochemical parameters and entropy. Table 1 shows the ranges of the parameters of the dataset. The input vector was represented by sequences with respect to HBV type. The primary database (DDBJ) that we used consisted of a set of oligo nucleotides from the literature. The oligos were compiled from published reports. The representational problem was addressed using different approaches, such as the definition and selection of physicochemical properties, the calculation of topological indices, and explicit vector-based representation of molecular connectivity. The exact number and type of descriptors that were used for a specific study were decided by an expert in the field.The encoding process required two subtasks to explicitly represent the relevant structural information in the molecules and to codify this structural information into a numerical representation.

 

Table 1. The statistical parameters of dataset and the MLR equation coefficients

Parameter

Statistics

Mean

Std Deviation

Variance

Skewness

Kurtosis

Minimum

Maximum

MLR Coefficient

aa Number

118.1904

23.30669

603.5575

-0.25747

-1.60215

83.7

156.6

-0.455

Molecular weight

13866.96

3108.255

10734724

-0.26862

-1.60774

9389.61

18367.92

0.005338

pI

6.327474

2.090067

4.853756

-0.3488

-1.46462

0.08316

8.361

-0.20592

Ala (A)

3.819443

1.55822

2.697833

0.153344

-0.584

0.9

7.83

0.028374

Arg (R)

7.018925

3.174779

11.19913

-0.12186

-1.4621

1.89

11.97

0.539712

Asn (N)

2.707171

0.969337

1.044014

-0.00856

-0.33279

0.9

5.85

0.083216

Asp (D)

5.155817

2.226123

5.506249

0.324161

-0.56204

1.71

10.98

0.321206

Cys (C)

6.723824

0.896244

0.892504

0.97551

0.231891

5.49

9.36

0.746696

Gln (Q)

5.68506

2.091403

4.859963

0.913855

0.517726

2.43

12.15

0.123994

Glu (E)

7.225817

1.849778

3.801865

0.485804

-0.54281

4.59

11.88

0.305088

Gly (G)

2.759881

1.082965

1.303124

0.940036

0.038038

1.17

5.49

0.515727

His (H)

3.064661

1.042359

1.207236

-0.30858

0.691933

0

6.21

-0.01709

Ile (I)

4.244702

0.903634

0.907283

-0.3925

-0.31292

1.71

6.75

-1.19643

Leu (L)

10.03518

1.478487

2.428803

0.180419

-0.42009

6.57

12.87

-1.29542

Lys (K)

4.255817

2.2008

5.381689

-0.4667

-0.91863

0

7.56

0.738991

Met (M)

1.322032

0.775334

0.667935

0.656632

-0.12146

0.54

4.32

0.571552

Phe (F)

2.290876

1.234493

1.693304

-0.15179

-0.23795

0

6.57

-0.30452

Pro (P)

4.148247

1.055127

1.236992

-0.05505

-1.05318

1.71

6.48

0.122032

Ser (S)

4.255099

1.257226

1.756241

0.955589

1.028613

1.26

8.1

0.064222

Thr (T)

5.714104

2.360066

6.188786

1.191432

1.110611

0.9

11.97

0.168044

Trp (W)

0.684143

0.686557

0.523733

0.596039

0.114345

0

3.15

0.413734

Tyr (Y)

3.921634

2.081592

4.814474

-0.20467

-0.7231

0

7.56

0.053179

Val (V)

5.03749

1.606633

2.868077

0.964922

1.658503

2.25

10.98

-0.74415

Asp + Glu

15.64064

2.045234

4.647758

-0.76209

-0.45523

10.8

18

-0.24517

Arg + Lys

16.00279

8.950353

89.00981

-0.22113

-1.61159

3.6

27

-0.56683

Sulfur

10.4522

2.286435

5.80865

0.85788

0.566699

6.3

16.2

-0.50919

Ext. coef.

14222.67

9224.946

94500000

0.385209

-0.66838

337.5

37606.5

-5.9E-05

instab. indx

54.08089

9.106119

92.13489

-0.42181

-0.31483

26.037

71.946

-0.00481

Aliph.

73.99043

7.536556

63.11075

0.412586

0.192556

59.976

95.904

0.360459

 

Constant

-25.486

 

Statistical Criteria
The statistical and graphical criteria were adopted to select the desired optimal network model. The selection was performed based on statistical accuracy measures, such as root mean square error (RMSE), coefficient of determination (R2), and mean absolute error (MAE). In the training and testing stages, these parameters were used to determine the most promising optimal network. The ideal values of RMSE and MAE are zero; a value near zero indicates a better-performing model. Values for R2 range from 0 to 1; higher values indicate better model agreement.

Network input, output, and preprocessing
After construction of the datasets that consisted of 700 amino acid sequences, they divided into two categories: resistant or sensitive to lamivudine. We assigned a value of 1 for resistance and 2 for sensitivity. Twenty-nine biochemical properties (Table 1), Shannon entropy, and different domains of each gene were calculated using available software mentioned above. All of the governed parameters were then normalized to between 0.1 to 0.8 and sorted randomly to reinforce the performance of the procedure under random conditions.

Model Development
The goal was to determine the relationship between chemiophysical features of mutant HBV YMDD and drug resistance. Construction of a training set that has patterns of a fixed length is difficult. A master list of chemiophysical features was compiled, including all features that appeared to be responsible for variations in contact point. Thus, the in put pattern that corresponds to a mutant has a pattern of features that leads to contact with the drug in a sensitive or resistant manner. The set of constructed patterns was then divided into a training set of 300 patterns and a validation set of 200 patterns. The training set included approximately 50% of the patterns with known resistance and 50% patterns without resistance values for a total of 300 training patterns. The validation set contained only patterns with known resistance and was used to test the prediction of the system. Finally, a network was trained with the patterns above.In the classical self-organizing feature map (SOFM) used here, all inputs were connected to all neurons. When a pattern is presented, the excitation of each unit is proportional to the dot product between the input vector and the weight vector. The unit with the weight vector that is closest to the input vector will have the largest excitation and will be declared the winner. The training involves changing the weights of the winner and its neighbors such that their weight vectors become more similar to the current input pattern. The training stops when the learning rate becomes zero. The experiments were performed with a learning rate of 0.6-0.9, and the learning rate was decreased linearly to zero over 10-50 training cycles.

Results

Using the collected dataset, a new model for predicting drug resistance in HBV was developed based on the ANN method. The results of the new MLP models are presented in Table 2 for the training, validation, and testing steps, and the statistical results of this model are presented in Table 3. The input parameters of this model are in Table 1, and the output parameter was HBV drug susceptibility.The ANN model extracted the dominant phenomena of resistance in the HBV RT gene and simulated its chemiophysical processes. Based on the results of Table 3, the R2, RMSE, and MAE values of the MLP in the training stage were 0.8834, 0.07, and 0.09, respectively, and 0.8465, 0.160.04 in the testing stage; based on the total data, the values were 0.8549, 0.115, and 0.065, respectively. In the MLR model, the values in the training stage were 0.799, 0.159, and 0.214, respectively, (Figure 1,2) and 0.4225, 0.9, and 0.26 in the testing stage, respectively; based on the total dataset, the values were 0.6107, 0.5295, and 0.237, respectively. From Table 3, it is clear that the MLP model is superior to the MLR model in predicting HBV lamivudine resistance. Also, in Table 3, the error prediction of the MLP and MLR models is shown. The results show that ANN models can be used as alternative methods for predicting the outcome of HBV therapy. The test results of different networks with regard to predicting drug responses are shown in Table 4.

Table 2. Test results for predicting drug responses

No.

Observed Output

MLP1 Prediction

MLP2 Prediction

Level

Error%

Level

Error%

1

2

1.72

22

1.96

17

2

2

1.72

22

1.96

17

3

2

1.72

22

1.96

17

4

2

1.96

17

2.08

21

5

2

0.68

81

0.32

98.6

6

2

1.04

63

2.24

27

7

1

1.3

45

1

25

8

1

1.6

75

1

15

9

1

1.6

75

1

15

10

1

1.6

75

1

15

11

1

1.6

75

0.9

15

12

1

1.2

35

1.9

25

13

2

1.8

20.5

1.9

15

14

2

1.7

26

1.9

17

15

2

1.7

24

1.9

18

16

1

0.9

19.4

0.9

16

17

1

0.8

26

0.9

18

18

1

0.9

25

0.9

18


Table 3.
Comparison of the MLP model in the testing stage with observed values

Model

Training

Testing

Total

R2

RMSE

MAE

Number of errors

R2

RMSE

MAE

Number of errors

R2

RMSE

MAE

Number of errors

MLP

0.8834

0.07

0.09

25

0.8465

0.16

0.04

21

0.8549

0.115

0.065

46

MLR

0.799

0.159

0.214

37

0.4225

0.9

0.26

31

0.6107

0.5295

0.237

68


Table 4.
Test results for different networks in prediction of oncogenicity

No

GF

MNN

RBF/GRNN/PNN

Level

Eror. %

Level

Eror. %

Level

Eror. %

1

2

17

2

15

2

19

2

2

17

2

15

2

21

3

2

19

2

15

2

21

4

2

41

2

19

2

21

5

1

99

1

87

1

35

6

1

89

1

77

2

19

7

1

15

1

15

1

35

8

1

55

1

16

1

35

9

1

65

1

16

1

35

10

1

65

1

16

1

35

11

1

25

1

15

1

35

12

2

15

1

35

1

45

13

2

16

2

15

2

30

14

2

15

2

16

2

24

15

2

15

2

16

2

31

16

2

15

2

16

2

37

17

2

16

2

15

2

26

18

2

15

2

16

2

33

 

 

 

Figure 1. Comparison of the MLP model in the training stage with the observed values

 

Figure 2. Comparison of the MLP model in the testing stage with observed values

Discussion

HBV drug resistance was predicted as a case study in this report. The proposed predictor showed vigorous clusterization of the predicted and observed drug responses. The current study was designed to find an algorithm that predicts drug resistance using chemiophysical information with artificially created neural networks. The applied properties for the model must be and rapid. After a pattern recognition model is constructed, the algorithm can be used to determine the repertoire of chemiophysical properties that is required for a drug response. This complex can be defined as the resistance pattern. This pattern can be used in a correlation analysis with known regulatory and biochemical molecular pathways to build a molecular model of drug resistance and response. Multifarious predictor models have been introduced for this purpose using different in vivo and in vitro architectures and parameters.
An MLP with one hidden layer and Tanh Axon as the transfer function can predict HBV drug resistance with an accuracy of 80% to 91%. Changes in the network structure, such as the addition of hidden layers, reductions in the threshold, alterations in the transfer function, randomization of data, and crossvalidation, did not improve the results, whereas in some cases, results with higher errors were obtained. Several other networks were used, some of which exhibited better results: generalized feed forward, fuzzy logic network, recurrent neural network, time-lag recurrent network, and modular neural network. However, the results from the multilayer perceptron network were most integral to rationale of error. In this study, the decision over the optimal structure was made on the basis of their compatibility with other applications and training time. However, decreasing the threshold 10-fold resulted in improved output. Ignoring a single ill-conditioned exemplifier, the error in the response of the rectified network ranged from 0% to 12%. Generalized feed forward, modular neural network, and RBF/GRNN/PNN are acceptable models of HBV drug response levels when trained with a 100-fold-decreased threshold and 20000 epochs.
Based on the data in Table 4, the performance of the networks can be classified as follows: the resistance level was predicted best by the multilayer perceptron and modular neural network, and the susceptibility level was predicted best by the multilayer perceptron, generalized feed forward, and modular neural network within the calculated ranges of error. Ignoring the single ill-conditioned test sample, the prediction by multilayer perceptron for two levels may be considered and for the final judgment on the response levels, other competent networks as described above should be consulted. In order to extend the objectives of the present study, an intelligent and multidisciplinary program should be developed based on information from different applications in similar investigations. This program aids in the automatic design of expert neural network architectures for each application.

References: (36)

  1. Mori K, Minami M, Kirishima T, et al. Prediction of Breakthrough Hepatitis due to Lamivudine-Resistant Hepatitis B Virus by a Sensitive Semi quantitative Assay Using Peptide Nucleic Acids. Intervirology. 2006;49 (5):274–80. [Link]
  2. Sablon E, Shapiro F. Advances in Molecular Diagnosis of HBV Infection and Drug Resistance. Int J Med Sci. 2005;2(1):8-16. [PubMed]
  3. Locarnini S, Mason WS. Cellular and virological mechanisms of HBV drug resistance. J Hepatol. 2006;44(2):422-31. [PubMed]
  4. Lo CM, Liu CL, Lau GK, Chan SC, Ng IO, Fan ST. Liver transplantation for chronic hepatitis B with lamivudine-resistant YMDD mutant using add-on adefovir dipivoxil plus lamivudine. Liver Transpl. 2005;11(7):807-13. [PubMed]
  5. Rajib H, Premashis K. HBV drug resistance: Its relevance in clinical practice. Hepat B Ann. 2007;4(1):24-39. [Link]
  6. Cao ZW, Han LY, Zheng CJ, et al. Computer prediction of drug resistance mutations in proteins. Drug Discov Today. 2005;10(7):521-9. [PubMed]
  7. Kiyosawa K, Tanaka E. Strategy for lamivudine-resistant YMDD mutant-associated chronic hepatitis B. J Gastroenterol. 2001;36(2):139-41. [PubMed]
  8. Yatsuji H, Noguchi C, Hiraga N, et al. Emergence of a novel lamivudine-resistant hepatitis B virus variant with a substitution outside the YMDD motif. Antimicrob Agents Chemother. 2006;50(11):3867-74. [PubMed]
  9. Sablon E, Shapiro F, Zoulim F. Early detection of hepatitis B drug resistance: implications for patient management. Expert Rev Mol Diagn. 2003;3(5):535-47. [PubMed]
  10. Ou ZY, Liu N, Chen CJ, Cheng G, He YS. Rapid and accurate genotyping of YMDD motif variants in the hepatitis B virus genome by an improved reverse dot blot method. J Clin Microbiol. 2005;43(11):5685-9. [PubMed]
  11. Cha CK, Kwon HC, Cheong JY, et al. Association of lamivudine-resistant mutational patterns with the antiviral effect of adefovir in patients with chronic hepatitis B. J Med Virol. 2009;81(3):417-24. [PubMed]
  12. Heo J, Cho M, Kim HH, et al. Detection of YMDD motif mutants by oligonucleotide chips in lamivudine-untreated patients with chronic hepatitis B virus infection. J Korean Med Sci. 2004;19(4):541-6. [PubMed]
  13. Stuyver L, Van Geyt C, De Gendt S, et al. Line probe assay for monitoring drug resistance in hepatitis B virus-infected patients during antiviral therapy. J Clin Microbiol. 2000;38(2):702-7. [PubMed]
  14. Whalley SA, Brown D, Teo CG, Dusheiko GM, Saunders NA. Monitoring the emergence of hepatitis B virus polymerase gene variants during lamivudine therapy using the LightCycler. J Clin Microbiol. 2001;39(4):1456-9. [PubMed]
  15. Ohishi W, Chayama K. Rare quasispecies in the YMDD motif of hepatitis B virus detected by polymerase chain reaction with peptide nucleic acid clamping. Intervirology. 2003;46(6):355-61. [PubMed]
  16. Hong SP, Kim NK, Hwang SG, et al. Detection of hepatitis B virus YMDD variants using mass spectrometric analysis of oligonucleotide fragments. J Hepatol. 2004;40(5):837-44. [PubMed]
  17. Bai YJ, Zhao JR, Lv GT, Zhang WH, Wang Y, Yan XJ. Rapid and high throughput detection of HBV YMDD mutants with fluorescence polarization. World J Gastroenterol. 2003;9(10):2344-7. [pubMed]
  18. Lin E, Hwang Y, Wang SC, Gu ZJ, Chen EY. An artificial neural network approach to the drug efficacy of interferon treatments. Pharmacogenomics. 2006;7(7):1017-24. [PubMed]
  19. Jajoo R, Mital D, Haque S, Srinivasan S. Prediction of Hepatitis C using Artificial Neural Network. Seventh lnternational Conference on Control, Automation, Robotics And Vision; 2002. New Jersey, USA: Control, Automation, Robotics and Vision; 2002. p. 1545-50. [Link]
  20. Maiellaro PA, Cozzolongo R, Marino P. Artificial neural networks for the prediction of response to interferon plus ribavirin treatment in patients with chronic hepatitis C. Curr Pharm Des. 2004;10(17):2101-9. [PubMed]
  21. Soleimanjahi H, Nategh MJ, Fakahi S, editors. A Performance Appraisal of Neural Networks Developed for Response Prediction across Heterogeneous Domains. World Academy Of Science, Engineering And Technology; 2009; Italy. [Link]
  22. Xiao Y, Segal MR. Prediction of genomewide conserved epitope profiles of HIV-1: classifier choice and peptide representation. Stat Appl Genet Mol Biol. 2005;4:Article25. [PubMed]
  23. Sinisi SE, Polley EC, Petersen ML, Rhee SY, van der Laan MJ. Super learning: an application to the prediction of HIV-1 drug resistance. Stat Appl Genet Mol Biol. 2007;6:Article7. [PubMed]
  24. Carvajal-Rodriguez A. The importance of bio-computational tools for predicting HIV drug resistance. Recent Pat DNA Gene Seq. 2007;1(1):63-8. [PubMed]
  25. Beerenwinkel N, Schmidt B, Walter H, et al. Diversity and complexity of HIV-1 drug resistance: a bioinformatics approach to predicting phenotype from genotype. Proc Natl Acad Sci U S A. 2002;99(12):8271-6. [PubMed]
  26. Douali L, Villemin D, Cherqaoui D. Exploring QSAR of Non-Nucleoside Reverse Transcriptase Inhibitors by Neural Networks: TIBO Derivatives. Int J Mol Sci. 2004;5(2):48-55. [Link]
  27. Wang D, Larder B. Enhanced prediction of lopinavir resistance from genotype by use of artificial neural networks. J Infect Dis. 2003;188(5):653-60. [PubMed]
  28. Draghici S, Potter RB. Predicting HIV drug resistance with neural networks. Bioinformatics. 2003;19(1):98-107. [PubMed]
  29. Rabinowitz M, Myers L, Banjevic M, et al. Accurate prediction of HIV-1 drug response from the reverse transcriptase and protease amino acid sequences using sparse models created by convex optimization. Bioinformatics. 2006;22(5):541-9. [PubMed]
  30. Bonet I, García M, Saeys Y, Van de Peer Y, Grau R. Predicting Human Immunodeficiency Virus (HIV) drug resistance using recurrent neural networks. In: Mira J, Álvarez J, editors. Bio-inspired Modeling of Cognitive Tasks: Springer Berlin / Heidelberg; 2007. p. 234-43. [Link]
  31. Wang D, Larder B. Enhanced prediction of lopinavir resistance from genotype by use of artificial neural networks. J Infect Dis. 2003;188(5 ):653-60. [PubMed]
  32. Yang ZR, Chou KC. Bio-support vector machines for computational proteomics. Bioinformatics. 2004;20(5):735-41. [PubMed]
  33. Terfloth L, Gasteiger J. Neural networks and genetic algorithms in drug design. Drug Discov Today. 2001;6(15 ):s102-s8. [Link]
  34. Beerenwinkel N, Lengauer T, Daumer M, et al. Methods for optimizing antiviral combination therapies. Bioinformatics. 2003;19(Suppl 1):i16-25. [PubMed]
  35. Amini-Bavil-Olyaee S, Hosseini SY, Sabahi F, Alavian SM. Hepatitis B virus (HBV) genotype and YMDD motif mutation profile among patients infected with HBV and untreated with lamivudine. Int J Infect Dis. 2008;12(1):83-7. [PubMed]
  36. Hosseini SY, Sabahi F, S A-B-O, Alavian SM, Merat S. A novel accurate ACRS-PCR method with a digestion internal control for identification of wild type and YMDD mutants of hepatitis B virus strains. J Virol Methods. 2006;137(2):298—303. [link]